The ServerFault stack exhange site has a tag named tcp-offload-engine: https://serverfault.com/questions/tagged/tcp-offload-engine

Practically all of the questions, however, are about partial forms of TCP offload. Not a single question is about TCP offload engine (the full form of TCP offload), although What problems can the use of TOE cause? comes close, but I believe the one who asked the question did not understand the difference between partial TCP offload (supported by Linux) and TCP offload engine (not supported by Linux).

The constant use of this tag for questions about partial TCP offload falsely lead people to believe that partial TCP offload would be similar to TCP offload engine. It is not! Partial TCP offload may in some cases be good because it uses all of the operating system TCP stack, whereas full TCP offload engine requires so major modifications to the operating system TCP stack that Linux will probably never support it. Of course some NIC manufacturer can hack together a Linux kernel version supporting TOE and release it under the GPL, but that will never, ever be accepted to the official kernel due to the reasons in the link I gave.

Hence, I propose that this tag would be renamed to tcp-offload. It is a proper name for questions about TCP offload engine (including the question I linked where the author probably didn't understand the difference between various forms of TCP offload), and also for questions about partial TCP offload such as send segmentation offload (gso, tso) and receive reassembly offload (lro, gro), and also for checksum offloads (rx, tx). If my prediction about the absurdity of full TCP offload engine is wrong, people can still continue to use the tcp-offload tag for TCP offload engine related issues.


Sounds reasonable to me. If there's no objections in the next few days, I'll make the change. Maybe remind me, @juhist, in case I forget. <grin>

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .